From bbesler@ouchem.chem.oakland.edu  Wed Jul 21 05:58:55 1993
Date: Wed, 21 Jul 93 09:58:55 -0400
From: bbesler@ouchem.chem.oakland.edu (Brent H. Besler)
Message-Id: <9307211358.AA10750@ouchem.chem.oakland.edu>
To: chemistry@ccl.net
Subject: Getting around G92 I/O bottleneck on RS/6000



I am configuring an RS/6000 cluster for an application(Gaussian 92)
which requires 2 Gigabyte scratch files(2 of them most of the  time
meaning 4 Gigabytes total) to be read in and out multiple times.  I/O
speed is the concern, data reliability not at all.  It seems that a 
level 0 RAID might solve the problem.  In some benchmarks which IBM did
for us they used a Zitel "cached" disk.  I am not sure if this is a RAID
or not?  Has anyone encountered this device before.  Also RAID's for the
IBM RS/6000 are mostly SCSI-2 interface.  There is a company which
produces RAID's for the RS/6000 called the MC Series which have a
proprietary Microchannel bus controller which is faster than SCSI-2 they
claim.  Has anyone ever used one before on an RS/6000? 
 I believe the comany which
manufactures them is Maximum Strategy Incorporated.  Also does anyone
know of a way around the 2.1 Gigabyte filesize limit of AIX? Finally
has anyone used Gaussian 92 with the scratch files going to NFS mounted
disks over an FDDI network?